I hope some of the mechanics here weigh in on this. 40 years ago I was working for a coffee company on a pouch pack packaging line. We had about a dozen machines that would put coffee in little mylar bags. I will just talk about the applicable part. A set of jaws would be going up and down on a carriage. They would grab the paper and while it was being pulled down it would fill with coffee, it would pull it down so far and then release. I was having a weird problem where with every stroke of the machine the bag would get a couple of thousands shorter. After about 10 minutes the bag would be too short and they would have to shut the machine off and immediately restart to fix the problem for another 10 minutes. The jaws were hydraulically activated by a solenoid controlled by a micro switch controlled by a spinning cam mounted on a 1/8" stainless steel shaft about 18" long, along with a few other cams. The problem would come and go and appear what seemed randomly on different machines. Now keep in mind that these machines had been upgraded to run about 50% faster than their factory limit. and were running flawlessly aside from this.
I will give folks a shot at this one before I give the answer, it had me stumped for 40 years.
By “paper” do you mean the mylar?
A diagram of the relevant parts would be useful though, in general, it sounds to me (a software engineer) like a synchronizing/measurement accuracy issue. For example, the computer thought the cam was spinning at 1000 rpm but when it’s running over-limits the ability to accurately measure the spin was wrong and, instead, it was 1004.1224675 rpm, which gradually caused it to fall of out sync between the two operations. Usually, you want to couple things together mechanically rather than using software to synchronize them or else rounding errors will cause a gradual mismatch.
If so, I’d guess that it always eventually falls out of sync but that, when under factory limits, the variance builds up so slow that the daily shutdown happens before the difference becomes notable.
Of course, as a software engineer, I’m going to think of the problem that would involve software. Maybe they were mechanically linked and it was a heat/warping sort of problem.
Most unusual. Most crossed index finger syndromes I’ve been involved with have the software and hardware engineers blaming the other. 
- I’m not on the project so it doesn’t hurt me to point at anyone in particular.
- Not my style. Whoever needs to fix it is the person who needs to fix it. Both sides should verify that their side isn’t the cause.
I thought this sounded familiar!
Haven’t you posted about this problem before? That it happens on different machines randomly is difficult to understand but you’ve provided very little information to work with. So if I’m guessing based on insufficient information I’ll go with the 18" long 1/8" shaft is undulating at the higher speed and cams are contacting micro-switches for the correct amount of time.
The “hydraulically activated” plus the overspeeding (why do people do that?) suggests to me that the hydraulic system wasn’t quite able to scavenge all the fluid on each stroke through the exit orifices in the reduced time available. So over time the driving cylinder became backlogged with stagnant fluid resulting in progressively increasing short-stroking.
The other likely possibility is a timing issue where the stroke starting time lagged just a bit vs the design point. And again in a cumulative fashion. Where in the electromechanical controls the excess delay was coming from is hard to say without a lot more detail. But for a system composed of microswitches, relays, and physical valves, it would be easy for there to be a certain minimum reaction time from “need to actuate X” to “X is actually actuated”. If that minimum time isn’t met, the actuation will be late. And if the return stroke is what does the timing reset, that will be late too. Cue accumulating error.
In fact it’s a good bet the rated speed of the machine is set to be “as fast as the control system can keep up without accumulating timing errors given normal tolerances”. So of course the OP’s employers turned the machine up to eleven and found it failing to keep its parts synced properly. Of course they did.
You nailed it! I had never considered the 18" shaft an issue at that time. I think centrifical force just continued to distort it at a very specific rate. My style of adjusting micro switches to the cams was to set them very sensitive when it really should have been pushed in a little further. I had never considered that either.
Yes, when I up graded the speed of the machine I should have also used a larger diameter shaft or shortened the anchor. point.
What was funny was in all the times I have described the problem to someone I never mentioned the 18"X1/8 shaft before because it never crossed my mind it was relevant. Within 10 seconds of me thinking about that shaft it hit me instantly.
Done the same thing myself so many times. It’s like an Agatha Christie mystery, there’s always a character that couldn’t possibly be the killer, but then it turns out they were. But in this case I have to say that it was an obvious weak link. Really, an 18" long 1/8" diameter shaft? Was this designed to work on the moon?
It was all mechanical with some unrelated electronics. RPM of the shaft was only 75 RPM,s
BTW, what does this mean? A couple of thousands what shorter?
It was a slow turning 75 RPM shaft with 1" little flat cams mounted on it. . Designed to be turning at 50 rpm. I think the biggest problem was my overly sensitive setting on the micro switches. Just a few thousands movement of the shaft would cause this problem.
Presumably thousandths. And the OP being a guy from the USA, presumably they’re thousandths of an inch.
Yes thousandths of an inch. centrifugal force slightly distorted it.
I’m not sure I believe this explanation.
Why would the error accumulate?
The centrical force of the spinning shaft would just gradually put a bow in the shaft that slowly increased as it spun.
I seriously doubt that. Besides that, where were the bearings supporting the shaft, and where was the cam located on it?
The bearing supports were about 18" apart and the cam in question was about 8" below the top bearing support.